AI-powered modeling of bee spermatozoa quality post agrochemical exposure

Abstract Understanding the impact of environmental stressors on reproductive health in insect pollinators is critical to addressing global biodiversity and food security challenges. Although artificial intelligence (AI) methods are increasingly applied to biological data, their use in modeling insect reproductive impairments remains limited. Here, we show that machine learning can reveal biologically meaningful fertility patterns in male (drone) honey bees ( Apis mellifera ) and enable interpretable predictions of pesticide-induced reproductive impairments. Using an integrated machine learning framework, we characterized spermatozoa quality in A. mellifera drones exposed to neonicotinoid pesticides. K-means clustering identified three distinct spermatozoa quality profiles (low, mid, high), which were validated against reproductive benchmarks and predicted with 96% accuracy using the Extreme Gradient Boosting (XGBoost) algorithm. SHapley Additive exPlanations (SHAP) analysis revealed live spermatozoa count as the most decisive predictor, underscoring the interpretability and biological relevance of the model. Our findings demonstrate that explainable AI can model pesticide-induced changes in male insect fertility, providing a scalable tool for ecotoxicological risk assessments as well as applications in honey bee queen breeding programs.